Allow for loading RF Diffusion embedding

FabianTraxler / ag_binding_affinity

Code repository for Master Thesis "Antibody-antigen binding affinity prediction through the use of geometric deep learning"

https://repositum.tuwien.at/bitstream/20.500.12708/188079/1/Traxler%20Fabian%20-%202023%20-%20Antibody-Antigen%20Binding%20Affinity%20Prediction%20through%20the...pdf

4 stars 0 forks source link

Allow for loading RF Diffusion embedding #14

Open MarcoMorik opened 1 year ago

MarcoMorik commented 1 year ago

Currently in /dataset/utils.py the load_graph_dict function can only load of_embeddings.

To train the affinity predictor with rf_diffusion embeddings, we need to:

[x] load precomputed RF diffusion embeddings (which could have the same format as the of embeddings)
[ ] Compute the RF diffusion embedding on the fly.

MihailBogojeski commented 1 year ago

Currently there seems to be a mixup in terms of how the OF and RF embeddings should be loaded, while ideally the Affinity Model should not care to distinguish between the two. Some mismatches about the approaches on loading these embeddings need to be resolved, specifically in load_graph_dict in src/abag_affinity/dataset/utils.py

moritzschaefer commented 1 year ago

Test how RF Diffusion gradient updates change a ground-truth protein (Marco)